CSE - Computing with Emerging Logic and Memory Technologies
There is ever growing interest in new memory technologies that can (1) expand the memory hierarchy by serving as storage-class memory between DRAM and solid-state drives (SSDs), (2) increase storage density owing to explosions in dataset sizes in AI, privacy preserving computing, etc., and (3) enable in situ computation to reduce overheads associated with data transfer – e.g., in-memory computing fabrics including (a) computing at the array periphery (CAP) that exploits the internal bandwidth of the memory to achieve parallelism/perform Boolean operations on different data words, (b) analog crossbar (xBar) arrays for highly parallel MVMs, and (c) content addressable memories (CAMs) that perform parallel search operations for data words in the memory per an input query and a desired matching function.
Numerous memory devices such as phase change memory (PCM), resistive random access memory (RRAM), spin devices, electrochemical memories, and ferroelectric devices continue to be studied in this space -- although it is still unclear whether or not they can deliver "wins" versus the existing technological and architectural state of the art.
Students will study whether or "wins" are possible when technology-enabled architectures are applied to "at scale" problems -- e.g., large language models (LLMs), privacy preserving computing and encryption, recommendation systems, scientific computing workloads, etc.
Our research is funded by the National Science Foundation, DARPA, and all major semiconductor/technology companies (e.g., Intel, TSMC, Samsung, IBM, etc.) Students will work on technologies and applications that the government and industry deem to be the most interesting/state-of-the-art. Prior students have also had engagements with personnel from said companies.